On the syntax of adversative coordination - Luis Vicente

Natural Language and Linguistic Theory manuscript No.
(will be inserted by the editor)
On the syntax of adversative coordination
Luis Vicente
Received: date / Accepted: date
Abstract A series of studies have distinguished two types of but, namely, corrective and
counterexpectational. The difference between these two types has been considered largely
semantic/pragmatic. This article shows that the semantic difference also translates into a
different syntax for each type of but. More precisely, corrective but always requires clauselevel
coordination, with apparent counterexamples being derived through ellipsis within the
second conjunct. On the other hand, counterexpectational but is not restricted in this way,
and offers the possibility of coordination of both clausal and subclausal constituents. From
this difference, it is possible to derive a number of syntactic asymmetries between corrective
and counterexpectational but.
Keywords Coordination · Ellipsis · Negation · Spanish · English
1 Introduction
This article examines the syntax of the adversative conjunction but, especially in its usage as
coordinator of (apparently) subclausal constituents. Although the literature on the syntax of
but is relatively small, 1 enough has been written to differentiate two competing proposals.
On the one hand, Sag et al. (1985), Bianchi and Zamparelli (2004), and Merchant (2004b)
argue that but can only coordinate clauses. 2 Under this analysis, apparent cases of but coordination
of subclausal constituents must be reanalyzed as clausal coordination plus ellipsis
within the second conjunct (1b). I’ll be assuming an analysis of ellipsis along the lines of
Luis Vicente
Department Linguistik, Universität Potsdam, Karl Liebknechtstraße 24-25, Golm 14476, Germany
E-mail: vicente@uni-potsdam.de
1 Most of the previous studies on but focus on its semantic and pragmatic aspects. See, e.g., Lakoff (1971),
Grice (1975), Anscombre and Ducrot (1977), Barwise and Cooper (1981), Horn (1989), Blackmore (1989,
2000), von Fintel (1994), Umbach (2005), and references therein. Since this is an article about the syntax
of but, I will not say anything about its semantics and pragmatics beyond the brief remarks in section 2,
and instead refer readers to the works just cited. Similarly, I won’t tackle non-conjunctive uses of but, such
as its use as an exceptive marker (e.g., the final sentence of the acknowledgements footnote), for which see
Reinhart (1991); or as a synonym for only (e.g., ‘He is but a man’).
2 The argumentation in this article doesn’t depend on adopting a specific definition of ‘clause’. For explicitness,
though, I’ll treat clauses as CPs whenever some left-peripheral position is required, and IPs otherwise.

2 Luis Vicente
Merchant (2001, 2004a) and related work—that is, in terms of movement of the remnant of
ellipsis to the left periphery plus PF deletion of IP. Throughout this article, elided material
is represented in a light grey font.
(1) a. Amanda ate three apples but one banana.
b. [Amanda ate three apples] but [[one banana] [ IP Amanda ate t]].
On the other hand, Barwise and Cooper (1981) claim that genuine DP-level but coordination
is possible, subject only to certain semantic restrictions. 3 No ellipsis is necessary
under this approach (2b).
(2) a. Amanda ate three apples but one banana.
b. Amanda ate [ DP three apples] but [ DP one banana].
This article shows that both analyses are correct: adversative coordination is compatible
with an analysis in terms of clausal coordination plus optional ellipsis (1b), as well as with
one in terms of small coordination without ellipsis (2b). However, and very importantly,
the choice between (1b) and (2b) is not random—rather, it is determined by the specific
semantic/pragmatic relation between the two conjuncts. The semantic literature has shown
that adversative coordination comes in two flavours, which have been traditionally called
corrective and contrastive—although, given that these two terms are somewhat similar and
therefore prone to confusion, I will relabel contrastive with counterexpectational. Corrective
but, corresponding to sino in Spanish, results in the denial of the proposition expressed in
the first conjunct.
(3) a. Amanda didn’t eat one apple but (rather) three bananas.
b. Amanda no comió una manzana sino tres plátanos.
Amanda not ate an apple but three bananas
On the other hand, counterexpectational but (pero in Spanish) does not deny the proposition
of the first conjunct. Rather, it simply compares two states of affairs, introducing the
implicature that the second conjunct is unexpected given the first conjunct (4).
(4) a. The girl is tall but no good at basketball.
b. La chica es alta pero desastrosa jugando al baloncesto.
the girl is tall but disastrous playing to.the basketball
We will see that the semantic/pragmatic difference between corrective and counterexpectational
but also has a reflection in syntax. Specifically, I defend the hypothesis that
corrective but (sino) always requires clause-level coordination, with an optional subsequent
step of ellipsis. On the other hand, counterexpectational but can directly coordinate subclausal
constituents (DPs, bare adjectives, etc.) without resorting to ellipsis. We will see in
the following sections that, from this asymmetry, it is possible to derive a number of syntactic
differences between the two types of but coordination—some of them unnoticed so far,
to the best of my knowledge. On a larger scale, we will also see that the data discussed here
3 Specifically, Barwise and Cooper argue that but requires that one conjunct be upward entailing and the
other downward entailing (in contrast with and, which they argue requires both conjuncts to be entailing in
the same direction). However, as one reviewer points out, this generalization is not as clear-cut as Barwise
and Cooper claim it is. For instance, in (1a) above, both conjuncts are upward entailing, yet the example is
perfectly well-formed.

On the syntax of adversative coordination 3
support the hypothesis that fragmentary sentences have a full (albeit silent) clause structure
(Morgan 1973, Merchant 2004a; and related literature), thus countering recent claims
to the contrary (Culicover and Jackendoff 2005, Stainton 2006, Nykiel and Sag 2009; and
references).
The article is organized as follows: in section 2, I elaborate a bit more on the semantic
differences between counterexpectational and corrective but. In section 3, I provide six arguments
that show that corrective but requires clausal coordination. After an intermediate
summary in section 4, section 5 applies the same arguments to counterexpectational but to
show that it allows coordination of a wider range of categories. Finally, section 6 summarizes
the empirical results, and section 7 offers a justification of why such an asymmetry
should exist.
2 Two types of adversative coordination
In this section, I introduce some notable characteristics of the two types of but, as a preparation
for the syntactic analysis to come. This section is not meant to be an exhaustive review
of their properties, and readers interested in the issues discussed here are instead referred to
the references in footnote 1.
2.1 Corrective but requires denial
One of the most notable characteristics of corrective but (sino in Spanish) is that the first
conjunct necessarily contains negation. Horn (1989:363ff) argues at length that what we
observe here is not a regular negation, but rather a metalinguistic negation, which he in turn
defines as “a device for objecting to a previous utterance on any grounds whatever, including
the conventional or conversational implicata it potentially induces, its morphology, its style
or register, or its phonetic realization”. This amounts to saying that corrective but (sino) is
used whenever we want to deny the proposition expressed by the first conjunct. The second
conjunct expresses a closely related, although true, proposition. It is the combination of the
denial of the first conjunct plus the assertion of the second that creates the corrective reading.
Horn (1989:397ff) also points out various differences between metalinguistic and regular
negation, concluding that they should be treated as separate phenomena. For instance,
metalinguistic negation cannot be incorporated into the morphology of a word in the clause,
which explains the ungrammaticality of (5a). Bosque (1980:137) observes the same restriction
for Spanish sino (5b). Only a morphologically independent, sentential negation licenses
corrective but (6). 4
(5) a. * This is improbable, but merely possible.
b. * Esto es improbable, sino meramente posible.
this is improbable but merely possible
(6) a. This is not probable, but merely possible.
b. Esto no es probable, sino meramente posible.
this is not probable but merely possible
4 Based on this restriction, Bosque (1980) suggests that sino is a negative polarity item. This conclusion,
however, is falsified by the data in (7) below, which show that the negation required to license sino cannot
license NPIs.

4 Luis Vicente
Similarly, both Horn (1989) and van der Wouden (1997:69) note that metalinguistic
negation cannot license negative polarity items. 5 As expected, clauses containing English
corrective but do not license NPIs, and Spanish sino does not license postverbal N-words. 6
Note that both examples in (7) are grammatical if ever and nunca are removed.
(7) a. I haven’t (*ever) been to Mexico but to Canada.
b. No he estado (*nunca) en México sino en Canadá.
not have been ever in Mexico but in Canada
2.2 Counterexpectational but introduces an implicature
In opposition to corrective but, counterexpectational but (pero) does not entail the denial of
the first conjunct. Rather, it simply gives rise to the implicature that the second conjunct is
somewhat unexpected given the first conjunct. Lakoff (1971) paraphrases this implicature
as “p (and therefore ¬q), but (actually) q” (see also Grice 1975 and related work). As an
illustration, consider the contrast below: (8a) is infelicitous because taxi drivers tend to
have driving licenses, so there is no sense in which the proposition expressed by the second
conjunct can be understood as unexpected. On the other hand, (8b) is more acceptable simply
because it is not normally expected of taxi drivers to hold truck driving licenses.
(8) a. # Randy is a taxi driver but he has a driving license.
b. Randy is a taxi driver but he has a truck driving license.
This property will affect the argumentation in section 5 slightly, in that the examples
there will have to be constructed in such a way that they satisfy the counterexpectationality
requirement. Counterexpectational but also differs from corrective but in the way it interacts
with negation. As (8b) also shows, counterexpectational but does not require the presence
of a sentential negation within the first conjunct. This is fully expected, as no denial of the
first conjunct is required in these cases. Furthermore, whenever such negation is present, it
can license NPIs and postverbal N-words unproblematically (9), suggesting that it is just a
regular negation, and not metalinguistic negation.
(9) a. I haven’t ever been to Mexico, but I have been to Canada.
b. No he estado nunca en México, pero he estado en Canadá.
not have been ever in Mexico but have been in Canada
2.3 The proposal
The thesis defended in this article is that the difference between corrective and counterexpectational
but goes beyond denial, counterexpectationality, and other semantic/pragmatic
aspects. Specifically, I will show that each type of but imposes different restrictions on the
syntax of their conjuncts, namely:
5 Thanks to Hilke Reckman (p.c.) for pointing this out to me.
6 Spanish is a non-strict negative concord language, which means that N-words (in this case, nunca) behave
differently depending on whether they appear in postverbal or preverbal position. When in postverbal
position, they need to be in the scope of a preverbal negative marker (typically, sentential negation), just like
English NPIs do; when in a preverbal position, however, they block the presence of sentential negation (see
Zeijlstra 2004 and references therein). In this example, nunca is in a postverbal position, hence it behaves
like an NPI in requiring the presence of a sentential negation.

On the syntax of adversative coordination 5
(10) The syntax of adversative coordination
a. Corrective but (sino) always requires its conjuncts to be full clauses.
b. Counterexpectational but (pero) allows its conjuncts to be smaller than clauses.
Two points merit further elaboration: First, note that the use of the words requires and
allows is not accidental. The claim here is that corrective but cannot coordinate anything
other than clause-level categories. In contrast, counterexpectational but is not restricted in
this way, and can coordinate constituents of any category, as long as the usual constraints
on unlike category coordination are respected (see Sag et al. 1985 and Munn 1993 for discussion).
This means that counterexpectational but can coordinate DPs, but also adjectives,
VPs, adverbs. . . and, crucially, also full clauses. Now, if we allow counterexpectational but
to coordinate full clauses, then we open up the possibility of ellipsis applying to second
conjunct, in the same way that I argue happens with corrective but coordination. We will
see in section 5 that this prediction is correct: When in an environment that independently
forces clause-level coordination (and, importantly, only in such an environment), counterexpectational
but starts exhibiting the same signs of ellipsis that are observed in corrective
but. I take this behaviour as strong evidence that the generalization in (10) is correct.
Second, (10) reduces all the syntactic differences between the two types of but to the
size of the conjuncts. It says nothing about the way in which the conjuncts combine with
the coordinator. The strongest way to interpret this conclusion is to say that both counterexpectational
and corrective but combine with their conjuncts in the same way. Here, I will
be assuming the asymmetric syntax for coordination defended in Munn (1993), Progovac
(1998a,b), and related works, where the coordinator is a head that takes the first conjunct as
its specifier and the second as its complement. 7 Graphically (and labeling the coordinator
‘&’ for simplicity):
(11) An asymmetric syntax for coordination
&P
CONJUNCT 1 & ′
& 0 CONJUNCT 2
The larger claim embodied here (and also implicit in Munn 1993) is that (11) is the
only possible syntax for all coordinators (and, or, corrective but, counterexpectational but).
This hypothesis entails that all syntactic asymmetries between coordinators stem not from
the syntax of the coordinate structure as a whole, but rather from the internal syntax of
their conjuncts. As just mentioned, this is something that follows from (10), which makes
reference exclusively to the category of the conjuncts.
7 Strictly speaking, Munn (1993) proposes that the coordinator plus the second conjunct form a maximal
projection (in his terms, a Boolean Phrase, or BP), which then right-adjoins to the first conjunct. Under this
analysis, the category of the whole coordinate structure is whatever the category of the first conjunct is. In
spite of this difference (which is not relevant for the purposes of this article), the constituency relations created
by Munn’s analysis are the same we observe in (11).

6 Luis Vicente
3 Corrective but requires clausal coordination
3.1 Scope of negation
The most obvious indication that corrective but requires a full clausal structure in its second
conjunct comes from examples like the following.
(12) a. Gabriel didn’t drink beer but champagne.
b. Gabriel no bebió cerveza sino champán.
Gabriel not drank beer but champagne
In both the English and Spanish versions of (12), the scope of negation is restricted to
the first conjunct only—i.e., they mean [(¬p) ∧ q]. Compare these examples with minimal
pairs where the coordinator is and (13), where a [¬(p ∧ q)] reading is possible. 8
(13) a. Gabriel didn’t drink beer and champagne.
b. Gabriel no bebió cerveza y champán.
Gabriel not drank beer and champagne
The source of this asymmetry can be traced to the requirements that and and corrective
but impose on their conjuncts. By hypothesis, corrective but can only coordinate clauses,
hence (12) must be assigned the structure in (14). In this structure, negation is embedded
inside the first conjunct. Hence, the fact that it cannot scope over the second conjunct follows
from a simple lack of c-command.
(14) The derivation of (13)
butP
CP but ′
Gabriel didn’t drink beer
but
CP
champagne
TP
Gabriel drank t
In contrast, and allows for coordination of smaller constituents without invoking ellipsis.
Example (13) can be assigned the structure in (15), where negation is outside the coordinate
structure. Therefore it can take wide scope over both conjuncts, yielding the [¬(p ∧ q)]
reading.
8 In reality, the examples in (13) are ambiguous between the [¬p ∧ ¬q] and the [¬(p ∧ q)] readings.
As one reviewer notes, the former reading can be forced by using the coordinator nor—see Repp (2005)
and Wurmbrand (2008) for discussion. The relevant point here is that the latter reading (equivalent, by de
Morgan’s Law, to [¬p∨¬q]), which requires negation scoping over coordination, is excluded from corrective
but coordination.

On the syntax of adversative coordination 7
(15) IP
Gabriel I ′
did
NegP
not
VP
drink
andP
beer and ′
and
champagne
3.2 Preverbal subject coordination
Corrective but cannot coordinate two preverbal subjects. 9 The examples in (16) cannot mean
‘two mathematicians got their papers published, but seven astrophysicists didn’t’, which
is the reading we would expect if it were possible to coordinate preverbal subjects with
corrective but.
(16) a. * Two mathematicians but seven astrophysicists didn’t get their papers published.
b. * Dos matemáticos sino siete astrofísicos no pudieron publicar sus
two mathematicians but seven astrophysicists not were.able publish their
artículos.
papers
In contrast, and can coordinate preverbal subjects without trouble. Note that, as discussed
in the previous section, and coordination differs from corrective but coordination in
placing both conjuncts under the scope of negation.
(17) a. ̌ Two mathematicians and seven astrophysicists didn’t get their papers published.
b. ̌ Dos matemáticos y siete astrofísicos no pudieron publicar sus
two mathematicians and seven astrophysicists not were.able publish their
artículos.
papers
This asymmetry can be explained if corrective but requires its conjuncts to be clauses,
while and allows DP coordination. It is not possible to analyze the sequence ‘two mathematicians
but seven astrophysicists’ in terms of clausal coordination plus ellipsis, given that
9 Sandra Chung (p.c.) and Jorge Hankamer (p.c.) have pointed out to me examples like (i), which seem to
falsify the claim I make in this subsection and the following one.
(i)
a. Not Steve but I should drive the car.
b. Not three but four girls are sunbathing on the lawn.
I’ll ignore such examples for the time being and return to them in section 3.7

8 Luis Vicente
the Backward Anaphora Constraint (BAC, see Langacker 1969, Ross 1967, 1969) prohibits
backward ellipsis within coordinate structures. 10
(18) a. * [Two mathematicians [ IP got their papers published]] but seven astrophysicists
didn’t get their papers published.
b. * [Dos matemáticos [ IP pudieron publicar sus artículos]] sino siete
two mathematicians were.able publish their papers but seven
astrofísicos no pudieron publicar sus artículos.
astrophysicist not were.able publish their papers
On the other hand, and allows direct coordination of the two subject DPs, thus circumventing
the restrictions of the BAC.
(19) a. ̌ [[ DP Two mathematicians] and [ DP seven astrophysicists]] didn’t get their
papers published.
b. ̌ [[ DP Dos matemáticos ] y [ DP siete astrofísicos ]] no pudieron
two mathematicians and seven astrophysicists not were.able
publicar sus artículos.
publish their papers
Note, however, that corrective but can coordinate right-peripheral subjects (20a). Obviously,
this effect is best illustrated in Spanish, since English doesn’t generally allow postverbal
subjects. 11 The grammaticality of this example is predicted by a conjunction reduction
analysis, since it allows a parse in which the second conjunct is part of an elided clause
(20b).
(20) a. ̌ No publicaron sus artículos dos matemáticos sino siete astrofísicos.
not published their papers two mathematicians but seven astrophysicists
‘Two mathematicians didn’t publish their papers but seven astrophysicists
did.’
10 More precisely, the original formulation of the BAC states that an anaphor cannot simultaneously command
and linearly precede its antecedent. This restriction is classically illustrated through the paradigm below
(from Ross 1969), on the assumption that it is the base position of the although clause that counts for purposes
of command.
(i) a. Although I don’t know who, I know he wants to see someone.
b. Although I know he wants to see someone, I don’t know who.
c. I know he wants to see someone, although I don’t know who.
d. *? I don’t know who, although I know he wants to see someone.
There is no operation that preposes a conjunct within a coordinate structure, analogously to the preposing of
the although clauses above. As a consequence, linear precedence and command go hand in hand in coordinate
structures, just as in (ic) and (id). Therefore, the ban against backward ellipsis in coordinate structures follows
as a corollary of the BAC. See Ross (1967:ch. 5) for additional discussion.
11 An anonymous reviewer points out that English allows postverbal subjects under locative inversion,
predicting that this construction will allow corrective but coordination of subjects. As an illustration of this
prediction, the reviewer offers (i). My informants actually reject (i), but no conclusions can be drawn from
this fact, given that they also reject the control example (ii).
(i)
(ii)
* Into the room didn’t run two clowns but (rather) three cowboys.
* Into the room didn’t run two clowns.

On the syntax of adversative coordination 9
b. [No publicaron sus artículos dos matemáticos] sino [[siete astrofísicos][ IP
publicaron sus artículos t]].
3.3 Attributive adjective coordination
In the same way as preverbal subjects, attributive adjectives cannot be coordinated with
corrective but: (21a) cannot mean ‘I didn’t read a short book, but I read a long one’. The
explanation is the same as in the previous section: The only way to derive (21a) out of a
clausal coordination structure would require an implausible combination of backward and
forward ellipsis (21b).
(21) a. * I didn’t read a short but long book.
b. * [I didn’t read a short book] but [I read a long book].
The problematic aspect of (21a) is the fact that it features backward ellipsis within a
coordinate structure, which is not a licit operation (see previous subsection). Spanish behaves
in the same way, although the data need to be constructed with some care. Due to the
fact that Spanish attributive adjectives are nearly always postnominal, 12 it is not possible to
construct an exact minimal pair to (21a) above. Nonetheless, it is possible to check for the
same effect by adding a PP to the right of the adjective (22a). The derivation of this example
would require backward ellipsis (22b), just as that of (21a). Therefore, its ungrammaticality
is expected.
(22) a. * Mauricio no ha leído un libro corto sino largo de Neal Stephenson.
Mauricio not has read a book short but long by Neal Stephenson
‘Mauricio hasn’t read a short book by Neal Stephenson, but he has read a
short one.’
b. * [Mauricio no ha leído un libro corto de Neal Stephenson] sino [Mauricio ha
leído un libro largo de Neal Stephenson].
The impossibility of attributive adjective coordination suggests that corrective but can
only coordinate full clauses. Consider now one additional paradigm in support of this hypothesis.
The problem with the examples above is that they require one step of backward
ellipsis. In principle, this problem could be circumvented by restricting all ellipsis to the
second conjunct. Interestingly, this strategy results in ungrammaticality if the remnant of
ellipsis is a bare adjective (23a).
12 Some adjectives, such as presunto ‘alleged’ or verdadero ‘true’ can be used prenominally (cf. Ticio
2003). As expected, they cannot be coordinated with sino:
(i)
* No han atrapado al verdadero sino presunto asesino.
not have caught to.the true but alleged killer
‘It’s not the true killer that has been caught, but the alleged one.’
Note, nonetheless, that the example becomes much better if the second adjective is introduced by its own
article al. I propose that such examples are cases of edge coordinations, as described in section 3.7 below,
and should therefore be analyzed along the lines discussed there.
(ii)
No han atrapado al verdadero sino al presunto asesino.
not have caught to.the true but to.the alleged killer
‘It is not the true killer that has been caught, but the alleged one.’

10 Luis Vicente
(23) a. * I didn’t read a short book, but long.
b. * I didn’t read a short book, but long [ IP I read [ DP a t book]].
(24) a. I didn’t read a short book, but a long one.
b. I didn’t read a short book, but [ DP a long one [ IP I read t].
The contrast between (23a) and (24) provides one further piece of evidence in favour
of a step of ellipsis in corrective but coordination. As is well-known, English doesn’t allow
extraction of an attributive adjective out of its containing DP (25a). In contrast, extraction of
a full DP is unproblematic (25a).
(25) a. * Long, I read [ DP a t book].
b. [ DP A long book], I read t.
We know that ellipsis can circumvent movement violations in some cases (Ross 1969,
Lasnik 2001), but there are also several cases where no rescuing effect is observed—see
Sauerland (1996), Merchant (2004a,b, 2008), and Vicente (2008) for discussion. On the
assumptions that (i) movement is an integral part of ellipsis, and (ii) the type of ellipsis
observed under corrective but coordination does not help rescue movement violations, 13 the
paradigm in (23) through (25) follows.
Spanish, however, behaves in a different way. As opposed to English, when ellipsis is
fully restricted to the second conjunct, both a bare adjective and a full DP are acceptable
ellipsis remnants.
(26) a. Mauricio no ha leído un libro corto sino largo.
Mauricio not has read a book short but long
‘Mauricio hasn’t read a short book but a long one.’
b. Mauricio no ha leído un libro corto sino
[largo] [ IP Mauricio ha leído [ DP un libro t]].
(27) a. Mauricio no ha leído un libro corto sino un libro largo.
Mauricio not has read a book short but a book long
‘Mauricio hasn’t read a short book but a long one.’
b. Mauricio no ha leído un libro corto sino
[ DP un libro largo] [ IP Mauricio ha leído t].
The problem lies in the fact that Spanish does behave like English in not allowing extraction
of attributive adjectives in non elliptical contexts.
(28) a. * Largo, Mauricio no ha leído [ DP un libro t].
long Mauricio not has read a book
‘Mauricio hasn’t read a long book.’
b. [ DP Un libro largo], Mauricio no lo ha leído t.
a book long Mauricio not CL has read
‘Mauricio hasn’t read a long book.’
13 Section 3.5 will provide additional evidence in favour of this assumption. For the time being, the reader
is referred to Kennedy and Merchant (2000) for an extensive study of the effects of ellipsis in attributive
adjective extraction.

On the syntax of adversative coordination 11
At this point, I am forced to say that whatever licenses adjective extraction under ellipsis
in Spanish is not operative in English. I do not have any deeper explanation, though, as to
why things ought to be this way. Nonetheless, the data in this subsection (especially (21)
through (22)) support the hypothesis that corrective but can only coordinate full clauses.
3.4 Agreement
As pointed out in section 3.2, corrective but can coordinate clause final subjects in Spanish.
This is because such examples offer the possibility of a parse in which the second conjunct
is actually part of an elided clause. This hypothesis makes an interesting prediction: Given
that the second conjunct belongs to a separate clause, it will not be able to trigger agreement
on the first conjunct verb. That is, in cases where corrective but conjoins two clause final
subjects, a first conjunct agreement effect arises. The following examples confirm this prediction.
14 Note that if the coordinator is and (instead of sino) we get regular full conjunct
agreement. This is because, as discussed above, and allows DP-level coordination, which
forces agreement with the whole coordinate structure.
(29) a. No se
not SE
{ ̌presentó / *presentaron
showed.up.3SG showed.up.3PL
} un
a
trombonistas.
trombone players
‘A pianist didn’t show up but three trombone players did.’
b. No
not
(30) a. No
not
se
SE
{ *presentó
showed.up.3PL
/ ̌presentaron
showed.up.3PL
} un
a
trombonistas.
trombone players
‘A pianist and three trombone players didn’t show up.’
{ ̌cometió / *cometieron
made.3SG made.3PL
} un error
a
pianista sino tres
pianist but three
pianista y tres
pianist and three
un pianista sino tres
mistake a pianist but three
trombonistas.
trombone players
‘A pianist didn’t make a mistake but three trombone players did.’
b. No { *cometió / ̌cometieron } un error un pianista y tres
not made.3SG made.3PL a mistake a pianist and three
trombonistas.
trombone players
‘A pianist and three trombone players didn’t make a mistake.’
To complete the argument, it is necessary to show that first conjunct agreement in Spanish
is really an illusion due to ellipsis, rather than a genuine first conjunct agreement effect
(for explorations of the latter option in various languages, see Johannessen 1998, van Koppen
2005, and references therein). The contrast between and and but illustrated in both (29)
and (30) already points towards this conclusion: If this was a genuine first conjunct agreement
effect, we wouldn’t expect it to be affected by the choice of coordinator.
14 As far as I know, this effect was first noted by Gallego (2004). However, he doesn’t attribute it to an
elliptical second conjunct. Rather, he assumes that there is no ellipsis and stipulates that coordinated subjects
in corrective but coordination behave as a “more compact unit” for purposes of agreement (Gallego 2004:20).

12 Luis Vicente
Furthermore, we can show that Spanish does not behave like languages that display genuine
first conjunct agreement. The hallmark of such languages is that first conjunct agreement
effects persist even when a clausal-coordination-plus-ellipsis analysis is otherwise impossible.
The two environments where this effect is typically exemplified are (i) modification
by together and (ii) binding of the reciprocal anaphor each other. The relevant feature
of together and each other is that they require the presence of a plural DP (cf. the ungrammaticality
of English *Gabriel came together and *Gabriel looked at each other). Consider
now the following examples from Standard Arabic (Soltan 2007), where the only pluralities
available are the coordinate subjects Hind and Amir and Hind and her brother. Crucially,
DP-level coordination is necessary in these two cases: a clausal-coordination-plus-ellipsis
alternative would generate two separate clauses, each with a singular subject unable to license
either together or each other on its own. Still, the verb shows singular agreement,
which suggests that it is agreeing exclusively with the first conjunct of a DP-coordination
structure.
(31) Standard Arabic
a. aa-t [ DP Hind-un wa amr-un ] maan.
came.3SG Hind.NOM and Amir.NOM together
‘Hind and Amir came together.’
b. Tuibbu [ DP Hind-un wa axaw-a-ha ] bad-a-hum el-bad.
love.3SG Hind.NOM and brother.NOM.her some.ACC.them the.some
‘Hind and her brother love each other.’
In Spanish, the modifier juntos ‘together’ and the reciprocal pronoun el uno al otro
‘each other’ behave like their Arabic counterparts in requiring the presence of a plural DP.
In the examples below, this plural DP is the coordinate structure Daniel y Gabriel. However,
unlike in Arabic, these environments do not license first conjunct agreement effects. The
ungrammaticality of first conjunct agreement in these examples can be explained if first
conjunct agreement effects in Spanish only arise in the context of clausal coordination plus
ellipsis. This context is bled in (32) by the presence of juntos and el uno al otro, hence the
necessity of full conjunct agreement.
(32) a. { ̌Vinieron / *vino } [ DP Daniel y Gabriel ] juntos.
came.3PL came.3SG Daniel and Gabriel together
‘Daniel and Gabriel came together.’
b. Se { ̌miraron / *miró } [ DP Daniel y Gabriel ] el uno al
SE looked.3PL looked.3SG Daniel and Gabriel the one to.the
otro.
other
‘Daniel and Gabriel looked at each other.’
In spite of this restriction, a first conjunct agreement effect can be observed with Spanish
and in cases where we can reasonably construct the second conjunct as belonging to a
separate, elliptical clause. This can be done by inserting a prosodic break (#) right before
the second conjunct and using the polarity particle tampoco ‘neither’. 15 Under this analysis,
(33a) is simply an elliptical variant of (33b).
15 For evidence that this construction does indeed involve clausal coordination plus ellipsis, see Depiante
(2000), Vicente (2006), and references therein.

On the syntax of adversative coordination 13
(33) a. No { ̌ha / *han } leído el libro Daniel, # y Gabriel tampoco.
not has have read the book Daniel and Gabriel neither
‘Daniel hasn’t read the book, and Gabriel hasn’t read it either.’
b. Daniel no { ̌ha / *han } leído el libro, y Gabriel tampoco lo ha leído.
Daniel not has have read the book and Gabriel neither CL has read
‘Daniel hasn’t read the book, and Gabriel hasn’t read it either.’
These data show quite clearly that Spanish first conjunct agreement effects are related
to the possibility of there being an elliptical clause. As a consequence, we can also conclude
that the obligatoriness of such effects under corrective but coordination indicates that
corrective but invariably requires clause-level coordination.
3.5 Locality effects
The theory of ellipsis I am assuming in this article is the one developed by Merchant (2001,
2004a,b), where the remnants of ellipsis move to a position outside the ellipsis site prior to
PF deletion. Given that movement is an integral part of this approach to ellipsis, we should
expect to find locality effects in cases of corrective but coordination. 16 This argument is
complicated, though, by the variable status of island effects under ellipsis. We can start
by noting that, while island effects disappear under sluicing (34), they persist in fragment
answers (35).
(34) a. ̌ They want to hire somebody who speaks a Slavic language, but I don’t know
which Slavic language.
b. * Which Slavic language do they want to hire somebody who speaks?
(35) A: They want to hire somebody who speaks Bulgarian.
B: * No, Polish.
B’: * No, Polish, they want to hire somebody who speaks.
The kind of ellipsis that I am hypothesizing for corrective but coordination can be considered
closer to fragment answers than to sluicing, given the non-interrogative status of the
second conjunct. Therefore, we should expect corrective but coordination to be sensitive to
island boundaries. 17 We will see in this section that this is indeed correct. 18 Let us begin by
noticing that some speakers judge sentences like (36), in which corrective but is coordinating
two objects, as degraded. Analogous examples in which corrective but coordinates two
subjects are judged as fully ungrammatical (38). The hypothesized extractions are illustrated
in (37) and (39).
(36) ?? I didn’t leave the party [{after/because} Amy started telling bad jokes] but [childhood
anecdotes].
16 To my knowledge, McCawley (1991) and Drubig (1994) were the first to discuss this effect.
17 Note that here it is important to place negation outside the island, so as to ensure that, because of parallelism,
the elided clause also contains an island boundary.
18 A few of the speakers I sampled (both for Spanish and English) do not find any island violations in the
relevant examples. I do not have anything interesting to say about this, other than speculating that, for these
speakers, ellipsis can circumvent island violations in a wider range of constructions than just sluicing (in fact,
Jason Merchant, p.c., informs me that a similar split in judgements can be observed with respect to examples
like 35). In spite of this, the fact remains that a significant subset of speakers agree with the judgements
indicated in the text, which shows that the island effects are real.

14 Luis Vicente
(37) I didn’t leave the party [{after/because} Amy started telling bad jokes] but
[childhood anecdotes] [I left the party [{because/after}Amy started telling t]].
(38) * I didn’t leave the party [{after/because} Amy started telling bad jokes] but [Cassandra].
(39) I didn’t leave the party [{after/because} Amy started telling bad jokes] but. . .
. . . [Cassandra] [I left the party [{after/because} t started telling bad jokes]].
As shown in (40) below, these judgements parallel the classical subject/object extraction
asymmetries (cf. Chomsky 1986 et seq). Hence I take the contrast in (36) vs. (38) to indicate
that corrective but involves movement as an integral part of ellipsis.
(40) a. ?? [Bad jokes], I left the party [{after/because} Amy started telling t].
b. * [Cassandra], I left the party [{after/because} t started telling bad jokes].
Note also that the examples in (36)/(38) become grammatical if corrective but doesn’t
coordinate only the objects/subjects, but rather the entire adjunct (41)/(43). This is because
movement of the adjunct doesn’t violate any constraints on movement (45). As above, the
(b) entries of the examples illustrate the corresponding movement operation.
(41) ̌ I didn’t leave the party [{after/because} Amy started telling bad jokes] but [{after/
because} she started telling childhood anecdotes].
(42) I didn’t leave the party [{after/because} Amy started telling bad jokes] but. . .
. . . [{after/because} she started telling childhood anecdotes] [I left the party t].
} {{ }
(43) ̌ I didn’t leave the party [{after/because} Amy started telling bad jokes] but [{after/
because} Cassandra started telling them].
(44) I didn’t leave the party [{after/because} Amy started telling bad jokes] but. . .
. . . [{after/because} Cassandra started telling them] [I left the party t].
} {{ }
(45) ̌ [{After/Because} Amy started telling bad jokes], I left the party.
As a final indication that movement is involved, note that the same speakers who reject
(36)/(38) find analogous examples without island boundaries much more acceptable (46).
(46) a. ̌ I didn’t say [that Cary Grant starred in The Rope] but [in Charade]. [cf. (36)]
b. ? I didn’t say [that Cary Grant starred in The Rope] but [James Stewart]. [cf.
(38)]
The following batch of examples show that the same paradigm can be replicated in
Spanish. In (47), we see there is an asymmetry as to whether sino coordinates two objects or

On the syntax of adversative coordination 15
two subjects embedded in an adjunct island; 19 in (48), we see that coordination of the entire
adjunct island circumvents ungrammaticality; and finally, in (49) we see that examples without
island boundaries are also grammatical. I do not provide derivations for these examples,
as they are structurally identical to the derivations for the English examples above.
(47) a. ?? Ernesto no se fue [{después de que / porque} Andrés empezara a
Ernesto not SE left after of that because Andrés started.SUBJ to
contar chistes malos] sino [batallitas de su infancia].
tell jokes bad but anecdotes from his childhood
‘Ernesto didn’t leave after/because Andrés started to tell bad jokes but childhood
anecdotes.’
b. * Ernesto no se fue [{después de que / porque} Andrés empezara a
Ernesto not SE left after of that because Andrés started.SUBJ to
contar chistes malos] sino [Mauricio].
tell jokes bad but Mauricio
‘Ernesto didn’t leave after/because Andrés started to tell bad jokes but after
Mauricio started to.’
(48) a. ̌ Ernesto no se fue [{después de que / porque} Andrés empezara a
Ernesto not SE left after of that because Andrés started.SUBJ to
contar chistes malos] sino [{después de que / porque} pro empezara a
tell jokes bad but after of that because started to
contar batallitas de su infancia].
tell anecdotes from his childhood
‘Ernesto didn’t leave after/because Andrés started to tell bad jokes but after/because
he started telling childhood anecdotes.’
b. ̌ Ernesto no se fue [{después de que / porque} Andrés empezara a
Ernesto not SE left after of that because Andrés started.SUBJ to
contar chistes malos] sino [{después de que / porque} Mauricio empezara
tell jokes bad but after of that because Mauricio started
a contarlos].
to tell.CL
‘Ernesto didn’t leave after/because Andrés started to tell bad jokes but after/because
Mauricio started telling them.’
(49) a. ̌ No he dicho
not have said
Charada].
Charade
[que
that
Cary Grant actuara
Cary Grant played.SUBJ
en La soga
in The Rope
], sino [en
but in
19 One potential problem with (47b) is that sino requires focus on the coordinated constituents, and for a
number of speakers there is a tendency to place focused subjects in a postverbal position. Thus, it is possible
that part of the deviance of (47b) is due to the placement of the subject in the first conjunct, rather than to
island constraints (thanks to Ricardo Etxepare, p.c., for pointing this out). Note that if we place the subject in
a postverbal position, we would expect a deviance similar to that of objects: postverbal subjects sit in SpecvP
(cf. Ordóñez 1997), which is a properly governed position. This seems to be correct.
(i)
?? Ernesto no se ha cabreado porque haya traído cinco suspensos su sobrino sino su hija.
Ernesto not SE has got.angry because has got five Fs his nephew but his daughter.
‘Ernesto didn’t get angry because his nephew got five Fs, but because his daughter did.’

16 Luis Vicente
‘I didn’t say that Cary Grant played in The Rope but in Charade.’
b. ? No he dicho [que Cary Grant actuara en La soga ], sino [James
not have said that Cary Grant played.SUBJ in The Rope but James
Stewart].
Stewart
‘I didn’t say that Cary Grant played in The Rope but James Stewart.’
In Spanish it is possible to use P-stranding as an additional movement test. 20 Example
(50a) shows that corrective but cannot coordinate two DPs under a single preposition—
rather, it is necessary for each conjunct to have its own preposition (50b). This follows if
corrective but requires clausal coordination plus movement as an integral part of ellipsis:
The derivation of (50a) requires movement of DP stranding its preposition, which is not
possible in Spanish (50c). The ungrammatical derivation that creates (50a) is illustrated in
(51). 21
(50) a. ?* No he visto a una chica con un vestido azul sino zapatos negros.
not have seen to a girl with a dress blue but shoes black
‘I haven’t seen a girl in a blue dress, but a girl in black shoes.’
b. ̌ No he visto a una chica con un vestido azul sino con zapatos negros.
not have seen to a girl with a dress blue but with shoes black
‘I haven’t seen a girl in a blue dress, but a girl in black shoes.’
c. * ¿Qué tipo de zapatos has visto a una chica con?
what type of shoes have seen to a girl with
‘What type of shoes have you seen a girl in?”
(51) * . . . sino [zapatos negros] [no he visto a una chica con t]
In short, we have seen that locality data point towards a clausal coordination analysis of
corrective but in which movement is an integral part of ellipsis.
3.6 Connectivity effects
Merchant (2004a) uses examples like the following to show that fragment answers stem
from a full clause that undergoes ellipsis.
(52) A: Who does every i man love the most?
B: His i wife.
The pronoun in (52B) gives rise to a bound variable reading, even though the clause it
appears in doesn’t contain any visible quantifier. Merchant explains this effect by assuming
that, in reality, (52B) contains an elided version of the quantifier. This results in a regular
quantifier-variable relation (53). If fragment answers were not elided clauses, the bound
reading of (52B) could only be explained by stipulating that variable binding can exceptionally
apply across utterances in these cases.
20 This argument is adapted from Depiante (2000:106ff), who uses it to show that regular stripping involves
A-bar movement.
21 See Rodrigues et al. (2009) and Vicente (2008) for arguments that, as opposed to what happens with
strong islands, ellipsis does not rescue P-stranding violations.

On the syntax of adversative coordination 17
(53) A: Who does every i man love the most?
B: [His i wife] [ every i man loves the most t].
The same argument can be made in the domain of corrective but coordination. Consider
the following pair of examples.
(54) a. I didn’t say that every i man loves his i wife but his i mistress.
b. No he dicho que todo i hombre quiera a su i mujer, sino a su i amante.
not have said that every man loves to his wife but to his mistress
These examples have two crucial properties we need to account for. First, as discussed
in section 3.1, the second conjunct in each example (his mistress/su amante) is interpreted
outside the scope of the negation in the first conjunct –i.e., these examples mean [(¬p) ∧
q]. Second, also in both cases, the only visible binder (the QP every man/todo hombre)
is contained inside the first conjunct, and therefore within the scope of negation. Suppose
now that, contrary to what I have been arguing so far, corrective but coordination allowed
DP-level coordination without ellipsis. If that were true, then (54) could have the following
structure (with irrelevant projections omitted for simplicity).
(55) A non-elliptical analysis of (54)
TP
I T ′
didn’t
VP
say
CP
that
TP
every man
TP
VP
loves
butP
his wife but ′
but
his mistress
This structure captures the fact that both his wife and his mistress are bound by a universal
quantifier, but it incorrectly derives a reading in which negation takes scope over
both conjuncts —i.e., [¬(p ∧ q)] instead of the actual [(¬p) ∧ q]. Note that this cannot be
remedied by raising his mistress to a position outside the scope of negation (ignoring, for
the sake of the argument, all the problems related to extraction out of a coordinate structure):
any position that is outside the scope of negation is also outside the scope of the only
universal quantifier in the tree. Therefore, trying to derive the correct scope of negation in

18 Luis Vicente
this way would cause the bound reading of his mistress to be lost. The conclusion is that a
DP-coordination analysis without ellipsis cannot simultaneously derive the correct scope of
negation and the bound reading of the second conjunct. 22
In contrast, the analysis is straightforward if (54) is a case of clausal coordination plus
ellipsis, analogous to Merchant’s analysis of (52). The structure corresponding to this kind
of analysis is given in (56). As we can see, the addition of silent structure places the second
conjunct outside the scope of the negation in the first conjunct. Moreover, the fact that the
second conjunct contains a universal quantifier of its own accounts for the bound reading of
his mistress. 23
(56) A clausal coordination analysis of (54)
butP
CP but ′
I didn’t say that
every man loves his wife
but
his mistress
CP
TP
I said that
every man loves [ t ]
We conclude that only a clausal coordination analysis can account simultaneously for
all the properties of (54).
3.7 An aside on edge coordinations
In sections 3.2 and 3.3, I argued that corrective but cannot coordinate preverbal subjects or
attributive adjectives. I attributed this restriction to the impossibility of creating the corresponding
elliptical structures. However, at the same time, I acknowledged in footnote 9 the
existence of cases like (57) and (58), which appear to contradict the claims defended in these
two sections. Bianchi and Zamparelli (2004) refer to such examples as “adjacent initial edge
coordinations”, though for convenience I’ll shorten the label to just “edge coordination”.
22 Note that one cannot get around this dilemma by raising his mistress out of the scope of negation and then
treating the bound reading of (54) as a case of donkey anaphora. Unlike in the examples in (54), a universal
quantifier requires a plural donkey-anaphoric pronoun, not a singular one (Evans 1980:341), both in English
and in Spanish:
(i)
a. If a farmer owns every donkey, he beats {̌them/*it}.
b. Si un granjero tiene todos los burros, pro {̌ les / *le } pega.
if a farmer owns all the donkeys CL.3PL CL.3SG beats
Similarly, one cannot assume that his mistress raises to escape the scope of negation and then reconstructs
exclusively to be bound by the universal quantifier. As Romero (1998) and Fox (1999) point out, reconstruction
appears to be an all-or-nothing type of operation: if we reconstruct for variable binding, then we are also
forced to reconstruct for scope, and we get back to the problem described in the paragraph after (55).
23 As pointed out to me by Jorge Hankamer (p.c.), one must assume that his mistress/su amante reconstructs:
otherwise, the movement proposed in (56) would take it outside the scope of its binder. The reader is
referred to Merchant (2004a) for discussion of how this analysis of ellipsis interacts with binding.

On the syntax of adversative coordination 19
(57) a. Not a mathematician but a physicist discovered the neutron.
b. Not three but four girls are sunbathing on the lawn.
(58) a. No un matemático sino un físico descubrió el neutrón.
not a mathematician but a physicist discovered the neutron
b. No tres sino cuatro chicas están tomando el sol en el jardín.
not three but four girls are taking the sun in the garden
There are reasons against conflating edge coordinations with the cases of preverbal subject
coordination discussed in section 3.2 (and, to a lesser extent, with the cases of attributive
adjective coordination in section 3.3). First, there is a very consistent word order difference:
edge coordinations exhibit the order [NEG DP BUT DP], whereas the order in preverbal subject
coordination and attributive adjective coordination is [DP BUT DP NEG]. Second, this
word order difference correlates with a difference in the scope of negation. In edge coordinations,
it is clear that negation takes scope only over the first conjunct. On the other hand, in
the cases of preverbal subject coordination, the intended scope of negation is over the second
conjunct only. These differences suggest that it is appropriate to treat edge coordinations as
a different phenomenon. The question, obviously, is what their proper analysis should be. At
first sight it might be tempting to conclude that edge coordinations feature corrective but coordination
at the DP/AP level, without ellipsis, which I have been claiming to be impossible.
The goal of this subsection is to address this potential problem for my analysis by showing
that edge coordinations are compatible with a clausal-coordination-only approach. 24
For a variety of reasons, most of the tests developed in sections 3.1 through 3.6 are
not applicable to edge coordinations. We can immediately dismiss tests based on subject
coordination and attributive adjective coordination, since these are the ones that give rise to
the problematic data. Tests based on connectivity effects are also useless here, as we will
see in section 5.6 that they cannot distinguish between clausal and subclausal coordination.
Of the three remaining tests, locality effects and scope of negation are also unreliable to
determine the correct structure. Let me start by commenting on locality effects on the basis
of the following example.
(59) * [Not the neutron but the Higgs boson] we had a toast because [a physicist had
discovered t].
Although it is obvious that (59) is an adjunct island violation, it is actually impossible to
tell what exactly causes the violation. We can say that edge coordinations are cases of small
coordination, and that the island violation is caused by movement of [ DP not the neutron
but the Higgs boson]. However, we could equally plausibly say that edge coordinations are
instances of clausal coordination plus ellipsis within the first conjunct, in which case the
ungrammaticality would be caused by the independent movements of not the neutron and
the Higgs boson. The conclusion is that, in this particular environment, locality effects are
also ineffective to differentiate between clausal and subclausal coordination.
A test based on the scope of negation is equally ineffective. Clearly, negation only scopes
over the first conjunct in (57) and (58). Note, however, that we are dealing here with constituent
negation: 25 as soon as we try to combine edge coordination with sentential negation,
we effectively replicate the ungrammatical sentences of sections 3.2 and 3.3.
24 This is also the hypothesis defended by Bianchi and Zamparelli (2004:326–327). See, however, Toosarvandani
(2009) for a dissenting view.
25 As one reviewer points out, this is quite transparent in languages like Greek, which feature different
lexical items for constituent and sentential negation (oxi and dhen, respectively). Although dhen is a verbal
clitic, I have also considered a sentence initial position in order to maintain a parallelism with oxi in (ia).

20 Luis Vicente
(60) a. * A mathematician but a physicist didn’t discover the proton.
b. * Three but four girls aren’t sunbathing on the lawn.
(61) a. * Un matemático
a mathematician
sino un físico no descubrió el neutrón.
but a physicist not discovered the neutron
b. * Tres sino cuatro chicas no están tomando el sol en el jardín.
three but four girls not are taking the sun in the garden
It is quite plausible to assume that constituent negation attaches directly to the negated
constituent (see Lasnik 1972 for English, and Depiante 2000 and Vicente 2006 for Spanish).
The problem is that this is again not enough to differentiate ellipsis from small coordination.
An example such as (57a) is potentially compatible with these two structures:
(62) Edge coordination as subclausal coordination without ellipsis
[ TP [[ DP not a mathematician] but [ DP a physicist]]
discovered the neutron].
(63) Edge coordination as clausal coordination plus ellipsis
[ TP [ DP not a mathematician] discovered the neutron]
but [ TP [ DP a physicist] discovered the neutron].
We are left, therefore, with only one test—namely, agreement, which fortunately offers
some clues about the correct structure of edge coordinations. The examples below show
that, when an edge coordination takes two singular DPs, the verb may only show singular
agreement. This is unexpected under a small DP coordination analysis, since a coordination
of two singular DP should be semantically plural. In contrast, this paradigm follows without
stipulation under a clausal coordination analysis. We may conclude, therefore, that edge
coordinations don’t constitute counterexamples to the analysis of corrective but coordination
developed above.
(64) a. Not a boy but a girl {*are/̌is} sunbathing on the lawn.
b. No un chico sino una chica { *están / ̌está} tomando el sol.
not a boy but a girl are is taking the sun
An additional argument in favour of this conclusion comes from the observation (Bianchi
and Zamparelli, 2004:314) that it is sometimes possible to separate the two conjuncts of an
edge coordination, giving rise to alternations like the one exemplified in (65). Any attempt to
derive (65b) via movement out of a small coordination structure would result in a violation
of the Coordinate Structure Constraint.
(65) a. [Not only Mary but also Lucy] did he invite.
b. [Not only Mary] did he invite, [but also Lucy].
(i)
a. ̌ Oxi tria
not three
b. * (Dhen) tria
not three
ala tessera koritsia kanun iliotherapia stin avli.
but four girls do heliotherapy in the yard
ala tessera koritsia
but four girl
(dhen)
not
kanun iliotherapia stin avli.
do heliotherapy in the yard

On the syntax of adversative coordination 21
The reader might have noticed that, even if edge coordinations are cases of clause-level
coordination, their surface form cannot be derived via ellipsis, as in the environments discussed
in sections 3.1 through 3.6. An analysis along these lines would require backward
ellipsis, and we have already seen in sections 3.2 and 3.3 that backward ellipsis is impossible
within coordinate structures due to the Backward Anaphora Constraint (Langacker
1969, Ross 1967, 1969). As an indication that edge coordinations require a reduction process
distinct from bona fide ellipsis, consider the fact that they disallow form mismatches
(cf. Bianchi and Zamparelli 2004, Gallego 2004). This is exemplified in (66) for number
morphology. In contrast, it is well-known that such mismatches are common under ellipsis,
as (67) illustrates. This difference suggests that, whatever the clause reduction process in
edge coordinations is, it is qualitatively different from ellipsis. The exact nature of this process
is not directly relevant to the point of this section (i.e., that edge coordinations involve
clausal coordination), so I will leave this aspect as an open question. 26
(66) a. ?? Not three but only one girl {are/is} sunbathing on the lawn.
b. ?? No tres sino sólo una chica {están / está} tomando el sol.
not three but only one girl are is taking the sun
(67) a. These women are more clever than Alfred [is clever].
b. Estas mujeres son más inteligentes que Alfredo [es inteligente].
these women are more intelligent than Alfred is intelligent
In conclusion, the data reviewed here show that edge coordinations involve clausal coordination
and, as a consequence, they do not constitute a counterexample to the hypothesis
defended in section 3. I appreciate that edge coordinations exhibit several other properties
that still need to be accounted for (see, e.g., the final sentence of the previous paragraph), but
that task is beyond the point of this section. Readers interested in a more in-depth analysis of
edge coordinations are instead referred to Bianchi and Zamparelli (2004), Gallego (2004),
and references cited there.
4 Interim conclusion and prospects
Table 1 summarizes the properties of corrective but. As we have seen in each of the six
cases, this particular array of characteristics follows from the hypothesis that corrective but
requires its conjuncts to be clauses.
Blocks scope of negation
Allows preverbal subject coordination
Allows attributive adjective coordination
Triggers first conjunct agreement
Shows locality effects
Shows connectivity effects
yes
no
no
yes
yes
yes
Table 1: Properties of corrective but
The argument will be completed in the next section, where we will see that counterexpectational
but behaves differently from corrective but in interesting ways. As already
26 Although one possible solution would be to adopt Hankamer’s (1973) proposal that apparent cases of
backward ellipsis within coordinate structures ought to be analyzed in terms of Right Node Raising.

22 Luis Vicente
advanced in (10b) above, the data will show that counterexpectational but allows coordination
of both clausal and subclausal constituents. In the cases where clausal coordination
takes place, we will observe the same signs of ellipsis discussed above for corrective but
coordination. On the other hand, subclausal coordination allows for a direct coordination
of the two relevant constituents, without any ellipsis taking place. Importantly, though, the
choice between these two types of coordination is not free. Rather, clausal coordination (plus
subsequent ellipsis) happens only when the second conjunct contains an element associated
to a high position in the clausal functional structure. In the data examined in section 5, this
element will be the negative particle no, which, after Laka (1990), Depiante (2000, 2004)
and Vicente (2006), I take to reside in a polarity projection ΣP in the extended CP area (I
return to this point below). In contrast, subclausal coordination appears to be the elsewhere
case. For ease of later reference, I summarize this split as in (68). The next section concentrates
on showing that (68) holds true. Discussion of the reason why it should hold true is
deferred to section 7.
(68) Behaviour of counterexpectational ‘but’
Counterexpectational but coordination is subclausal coordination without ellipsis
unless the second conjunct contains an element associated to the CP area. In the
latter case, clausal coordination plus ellipsis obtains.
Before proceeding to the next section, it is necessary to say something about the negative
particle no, which I will use to motivate a clausal coordination analysis. More specifically,
the relevant environment is the polarity ellipsis construction, exemplified in the second conjunct
below.
(69) Esteban ha viajado a Venezuela, pero a
Esteban has traveled to Venezuela but to
Cuba no.
Cuba not
Depiante (2000, 2004) and Vicente (2006) argue that such examples are derived by
moving the DP remnant to a topic position to the left of negation, followed by deletion of
IP (70). I take the negative marker no to head the polarity projection ΣP proposed in Laka
(1990). 27 I refer the interested readers to the cited papers for full justification of this analysis.
(70) . . . pero a Cuba [ ΣP no [ IP ha viajado Esteban t]].
It is worth noting that the construction exemplified above has a variant in which the
polarity marker precedes the remnant of ellipsis—i.e., a no XP order. The difference between
the two variants, however, is deeper than just a change in word order. Concentrating
especially on negative fragments, Depiante (2000, 2004) presents evidence that no XP orders
are arguably cases of constituent negation—i.e., no left-adjoins directly to the negated
constituent, without any ellipsis happening.
27 This assumption receives support from French, where the negative word used in polarity ellipsis is non,
distinct from the regular sentential negation ne. . . pas. Thanks to Amanda Morris (p.c.) and an NLLT reviewer
for pointing out this datum to me.
(i)
Marie a lu un livre, mais Claude non.
Marie has read a book but Claude not

On the syntax of adversative coordination 23
(71) a. Esteban ha viajado a Venezuela, pero no
Esteban has traveled to Venezuela but not
b. . . . pero [ PP no [ PP a Cuba]].
a Cuba.
to Cuba
Due to this asymmetry, I will consider only the XP no order in the next section, as this
is the only one that could potentially create an underlying clausal structure for the second
conjunct. Note also that the not XP order is the only possibility in the English polarity ellipsis
construction (72).
(72) a. Steve wants to go to California, but not to Vermont.
b. * Steve wants to go to California, but to Vermont not.
Lasnik (1972) claims that English polarity ellipsis is structurally ambiguous: it may
stem from either a conjunction reduction structure, as Spanish (70), or from non-elliptical
structure like Spanish (71). Given that it is not possible to guarantee that an English not
XP structure stems from an underlying clause, English data will play a smaller part in the
discussion in section 5. Spanish data will nonetheless suffice to show that (68) holds.
5 Counterexpectational but allows subclausal coordination
5.1 Scope of negation
As opposed to corrective but, counterexpectational but (Spanish pero) allows a negation
to take scope over both conjuncts, yielding the reading [¬(p ∧ q)]. Thus, the meaning of
(73a) is that it is not the case that Susie is simultaneously poor and honest, though she
might have one of these two properties (given that, by de Morgan’s law, [¬(p ∧ q)] = [¬p ∨
¬q]). As discussed above in section 3.1, this particular reading is generated when negation
scopes over the whole coordinate structure. Therefore, its availability supports an analysis in
which pero conjoins two bare APs, without any ellipsis. Compare this to the corrective but
coordination in (73b), which, as discussed in section 3.1, only allows a [(¬p) ∧ q] reading
as a consequence of being based on clausal coordination plus ellipsis. 28
(73) a. Susana no es [ AP pobre
Susana not is poor
[¬(p ∧ q)]
b. [ CP Susana no es pobre] sino
Susana not is poor but
[(¬p) ∧ q]
pero honesta].
but honest
[ CP honesta].
honest
28 Note that DPs exhibit the same behaviour as APs in this environment. Thanks to John Moore (p.c.) for
suggesting the following example.
(i)
a. Mario no es una persona generosa pero un mentiroso.
Mario not is a person generous but a liar
‘Mario is not both a generous person and a liar” [¬(p ∧ q)]
b. Mario no es una persona generosa sino un mentiroso.
Mario not is a person generous but a liar
‘Mario is not a generous person, but he is a liar” [(¬p) ∧ q]

24 Luis Vicente
Crucially, note that [¬(p ∧ q)] is the only possible reading of (73a). This is important in
that it confirms the hypothesis in (68) that clausal coordination under counterexpectational
but only occurs when forced by the presence of a CP-level element, with subclausal coordination
being the elsewhere case. To understand this, suppose that (68) didn’t hold and that
both clausal and subclausal coordination were equally available. If this were the case, then
we would expect (73a) to be ambiguous between the [(¬p) ∧ q] and the [¬(p∧q)] readings.
The fact that (73a) is not ambiguous in this way suggests that clausal coordination (i.e., the
syntax that generates the [(¬p) ∧ q] reading) is not freely available under counterexpectational
but coordination—rather, it only arises when required by the presence of a specific
element such as no.
Consider now how clausal coordination can be forced. Example (74) is a case of counterexpectational
but coordination with the polarity particle sí ‘yes’ in the second conjunct
(after Laka 1990, I assume that sí is an affirmative instantiation of Σ). The reading of this
example is [(¬p) ∧ q], which requires negation to scope over the first conjunct only. By
analogy with (73b), we can conclude that the derivation of (74) requires clausal coordination
plus ellipsis. As advanced in (68) above, this derivation is enforced by the presence of
a polarity particle in the second conjunct.
(74) Susana no ha viajado a Venezuela, pero a México sí.
Susana not has traveled to Venezuela but to Mexico yes
‘Susana hasn’t traveled to Venezuela, but she has traveled to Mexico.’
English is less informative than Spanish in this respect, given that it features one single
lexical item (but) for the two types of adversative coordination. Therefore, the following
example is ambiguous between the [(¬p) ∧ q] reading (when but is interpreted correctively)
and the [¬p ∧ ¬q] reading (when it is interpreted counterexpectationally). 29
(75) Susan is not poor but honest.
Ambiguous: [(¬p) ∧ q] and [¬(p ∧ q)]
5.2 Preverbal subject coordination
As opposed to corrective but, counterexpectational but can coordinate preverbal subjects
without trouble. In the same way as in the previous subsection, the grammaticality of (76)
follows if counterexpectational but and pero allow DP-level coordination without ellipsis.
(76) a. ̌ [ DP One single neurosurgeon but at least three cardiologists] will take part in
this operation.
b. ̌ [ DP Un único neurocirujano pero al menos tres cardiólogos]
one single neurosurgeon but at least three cardiologists
participarán en esta operación.
take.part.FUT in this operation
Also, as in the previous section, the polarity ellipsis construction cannot be used to
coordinate preverbal subjects. This suggests that, in the same way as with corrective but,
we are dealing with an elliptical clause here. Note that this sentence is grammatical if the
second conjunct appears at the right edge, in a position consistent with conjunction reduction
29 Although the speakers I have consulted tend to treat the [(¬p) ∧ q] reading as the primary one, with the
[¬(p ∧ q)] reading requiring some contextual prompting.

On the syntax of adversative coordination 25
(77b). 30 This shows that the ungrammaticality of (77a) is purely syntactic and cannot be
reduced to semantic incongruity.
(77) a. * [Un neurocirujano pero (al menos) tres cardiólogos no] participarán en
one neurosurgeon but at least three cardiologists not take.part.FUT in
esta operación.
this operation
‘One single neurosurgeon will take part in this operation, but (at least) three
cardiologists will not.’
b. Un neurocirujano participará en esta operación, pero tres cardiólogos no.
a neurosurgeon take.part.fut in this operation but three cardiologists not
‘One single neurosurgeon will take part in this operation, but (at least) three
cardiologists will not.’
5.3 Attributive adjective coordination
As an extension of the previous argument, consider the fact that counterexpectational but
also allows coordination of attributive adjectives (as opposed to corrective but, see section
3.3). Once again, the reason is that counterexpectational but allows its conjuncts to be
smaller than clauses (in this case, bare adjectives).
(78) a. ̌ A [ AP young but brilliant] organist played a Bach sonata.
b. ̌ Un organista [ AP joven pero brillante] interpretó una sonata de Bach.
a organist young but brilliant played a sonata by Bach
Also as in the previous section, trying to apply the polarity ellipsis construction to counterexpectational
but coordination of attributive adjectives results in ungrammaticality, unless
it is done in such a way that it becomes compatible with a conjunction reduction analysis.
(79) a. * Un organista [ AP joven] pero [ CP brillante no] interpretó una sonata de Bach.
a organist young but brilliant not played a sonata by Bach
b. Un organista [ AP joven] interpretó una sonata de Bach, pero uno brillante no.
a organist young played a sonata by Bach but one brilliant not
5.4 Agreement
In section 3.4, we saw that, in Spanish, corrective but is exceptional in triggering first conjunct
agreement with clause final subjects. That effect was attributed to the fact that the
second conjunct belongs to a separate clause and, as such, it is unable to trigger agreement
on the first conjunct verb. Example (80) below shows that the first conjunct agreement effect
disappears with counterexpectational but. This judgement is expected under the hypothesis
defended in this section that counterexpectational but can coordinate subclausal constituents
without any ellipsis taking place.
30 Note, though, that (77a) has an alternative grammatical parse, namely ‘One single neurosurgeon but at
least three cardiologists won’t take part in this operation’, where no is a sentential negation that doesn’t form
a constituent with the subject. This reading is irrelevant for our purposes, since we are focusing on cases
where negation only modifies the second conjunct.

26 Luis Vicente
(80) {̌ Van
go.3PL
/* va
go.3SG
} a
to
pero tres cardiólogos].
participar en la operación
take.part in the operation
[ DP un único
a single
neurocirujano
neurosurgeon
but three cardiologists
‘Although only one neurosurgeon will take part in the operation, three cardiologists
will.’
It is important to note that (80) only allows a full conjunct agreement effect. 31 This
restriction supports the hypothesis in (68) that, while counterexpectational but can coordinate
both clausal and subclausal constituents, both options are not equally freely available. If
they were, we would expect (80) to be grammatical with both first conjunct and full conjunct
agreement. The fact that the former option is ungrammatical suggests that clausal coordination
is not possible in this particular example. In other words, clausal coordination (and the
subsequent first conjunct agreement effect) is a marked option, with subclausal coordination
being the elsewhere case.
As discussed in the previous subsections, the way to enforce counterexpectational but
to coordinate full clauses is by combining it with the polarity ellipsis construction. As (81)
shows, judgements are reversed in this environment, and only first conjunct agreement is
possible.
(81) { *Van / ̌Va } a
go.3PL go.3SG to
tres cardiólogos
pero
but
three cardiologists
no.
not
participar en la operación
take.part in the operation
un único
a single
neurocirujano
neurosurgeon
5.5 Locality effects
In section 3.5 we saw that, for some speakers, corrective but is sensitive to island boundaries.
This sensitivity was attributed to the fact that movement is an integral part of the analysis
of ellipsis assumed throughout section 3. Those same speakers, however, find it perfectly
grammatical to embed a counterexpectational but coordinate structure inside the same type
of island (82). These examples shows that counterexpectational but does not involve ellipsis,
at least inasmuch as we want to consider movement an integral part of ellipsis.
31 While the judgements in (80) are pretty robust, I found that a first conjunct agreement is marginally
acceptable for some speakers, as long as a heavy prosodic break separates the two conjuncts (i). Given that
this effect is very weak, I do not consider it a significant problem for the discussion in the main text.
(i)
?? Va a participar en la operación un único neurocirujano # pero tres cardiólogos.
go.3SG to take.part in the operation a single neurosurgeon but three cardiologists
‘Although only one neurosurgeon will take part in the operation, three cardiologists will.’
Arguably, the status of (i) is due to the fact that, because of the clause-final placement of the subjects, this
example is actually compatible with an elliptical analysis. Compare it to (ii), where ellipsis is not an option
due to the preverbal position of the subjects (cf. sections 3.2 and 5.2): here, a full conjunct agreement effect
is unanimously required.
(ii)
Un único neurocirujano pero al menos tres cardiólogos
one single neurosurgeon but at least three cardiologists
la operación.
the operation
{ ̌van / *va } a
go.3PL go.3SG to
participar
take.part
en
in

On the syntax of adversative coordination 27
(82) a. ̌ I complained to the director of the hospital [because one single surgeon but
at least three unqualified students took part in the operation].
b. ̌ Me quejé al director del hospital [porque un único
CL complained to.the director of.the hospital because one single
neurocirujano pero al menos tres estudiantes sin experiencia
neurosurgeon but at least three students without experience
participaron en la operación].
took.part in the operation
However, island effects reappear if the second conjunct features polarity ellipsis. As discussed
in section 4, the string tres estudiantes no should be analyzed as part of an elliptical
clause, with movement (and therefore island sensitivity) being an integral part of ellipsis.
The derivation of the second conjunct of (83) is given in (84).
(83) * Me quejé al director del hospital [porque un único neurocirujano
CL complained to.the director of.the hospital because one single neurosurgeon
había participado en la operación] pero [tres estudiantes no].
had taken.part in the operation but three students not
(84) . . . pero [[tres estudiantes] no [ IP me quejé al director del hospital [porque t no habían
participado]]].
The same asymmetry holds for P-stranding effects: under regular counterexpectational
but coordination, they are absent, which suggests DP coordination below the preposition
(85a). However, if polarity ellipsis forces the second conjunct to be clausal, P-stranding
effects reappear. As above, the derivation of the second conjunct of the ungrammatical (85b)
is given in (86)
(85) a. ̌ He visto a una chica [ PP con [ DP un vestido azul pero zapatos negros]].
have seen to a girl with a dress blue but shoes black
‘I have seen a girl wearing a blue dress but black shoes.’
b. * He visto a una chica [ PP con un vestido azul] pero [ CP zapatos negros
have seen to a girl with a dress blue but shoes black
no].
not
‘I have seen a girl wearing a blue dress, but not black shoes.’
(86) . . . pero [[zapatos negros] no [ IP he visto a una chica [ PP con t]]].
5.6 Connectivity effects
Connectivity effects are the one point where counterexpectational but shows the same behaviour
as corrective but: a pronoun in the second conjunct can be bound by a preceding
quantifier.
(87) a. Every i medieval king was cruel towards his i servants but very caring for his i
horse.

28 Luis Vicente
b. Todo i rey medieval era cruel con sus i vasallos pero cariñoso con su i
every king medieval was cruel with his servants but caring with his
caballo.
horse
This possibility is expected if counterexpectational but allows for small coordination. If
only the objects are coordinated in (87), then the quantificational subject takes scope over
the entire conjunction, and can bind the pronouns in either object.
(88) IP
DP I ′
every i king was but P
AP but ′
cruel towards
his i servants
but
AP
caring for
his i horse
(89) Todo i rey medieval se
every king medieval SE
su i caballo no.
his horse not
comportaba
behaved
de manera
of manner
cruel con sus i
cruel with his
vasallos,
servants
pero con
but with
The data in (87) are compatible both with a small coordination analysis as well as with
a clausal coordination analysis in which the second conjunct contains a silent instance of
the quantifier (cf. section 3.6). As such, they do not constitute direct evidence in favour of
a small coordination analysis of counterexpectational but—they are rather merely compatible
with it. However, when this paradigm is considered together with the preceding five
arguments, a small coordination approach is more plausible than conjunction reduction.
6 Final empirical summary
The empirical results of this article are summarized in Table 2 below. As we have discussed
all through the article, these properties can be derived if we accept the following
three premises:
– Corrective but requires clausal coordination in all cases.
– Counterexpectational but allows coordination of both clausal and subclausal constituents.
– Notwithstanding the last point, counterexpectational but will show signs of clausal coordination
plus ellipsis only when the second conjunct contains an element associated
to the CP area (in the cases examined here, this element is the negative particle no in the
polarity ellipsis construction). In the elsewhere case, subclausal coordination without
ellipsis obtains.

On the syntax of adversative coordination 29
Corrective but
Counterexpectational but
(polarity ellipsis)
Counterexpectational but
Blocks scope of negation yes yes no
Allows preverbal subject coordination no no yes
Allows attributive adjective coordination no no yes
Triggers first conjunct agreement yes yes no
Shows locality effects yes yes no
Shows connectivity effects yes yes yes
Table 2: Comparison of corrective vs. counterexpectational but
(elsewhere)
7 Additional theoretical remarks
Table 2 shows that clausal coordination is required for corrective but coordination in every
case, and for counterexpectational but coordination when combined with the polarity ellipsis
construction. The common characteristic of these two environments is that they contrast
two propositions of opposite polarity. In contrast, in counterexpectational but coordination
outside polarity ellipsis environments, polarity is not an issue. In this case, what is contrasted
is a subconstituent of the clause—specifically, in the cases we have discussed, a DP or an
adjective. I want to propose that the data in Table 2 are a consequence of this asymmetry.
I start by assuming that polarity is encoded in a functional projection in the expanded
CP area –see Cinque (1999) and related works for discussion. In keeping with the terminology
of section 4, I will use Laka’s (1990) ΣP label to refer to this projection. With this
much in place, the data in Table 2 can be derived if the presence of ΣP entails the presence
of a full clausal structure embedded under it. I’ll turn to the justification of this assumption
in a moment, but first consider the reasoning behind it: corrective but coordinates a negative
proposition (technically, the denial of a proposition) with a closely related affirmative
proposition. Assuming that negation and affirmation require the presence of ΣP (pace Laka
1990), we can restate the previous sentence by saying that corrective but coordinates two
constituents of ΣP category. Now, if, as conjectured above, ΣP entails the presence of a full
clausal structure embedded under it, the fact that corrective but can only take full clauses as
its conjuncts follows. The same reasoning holds for counterexpectational but in combination
with polarity ellipsis. In this case we are also conjoining an affirmative proposition with a
negative one—therefore, two ΣP constituents, each containing a full clausal structure.
Let me turn now to the question of why the presence of ΣP should entail the presence
of a full clausal structure. This correlation can be derived from the selectional properties
of lexical items. That is, Σ must take a TP 32 complement to satisfy its subcategorization
feature. In turn, T must take a complement of the appropriate category (say, AspP), which
32 Obviously, this is only for the sake of exposition –whether TP is the immediate complement of Σ, or
whether there are intervening categories is irrelevant for the discussion. The same holds for the following
sentence.

30 Luis Vicente
also in turn must take a complement of its own, and so on. The necessity of satisfying this
sequence of selectional requirements gives us the desired result.
Crucially, this is not the case when counterexpectational but is not associated with polarity
ellipsis: in such examples, the polarity of the conjuncts is not contrasted, therefore
there is no need for an independent ΣP in each conjunct. Consequently, there is no need either
for a full clausal structure in each conjunct, and coordination of subclausal constituents
obtains. Note that the reasoning outlined here depends on the assumption that clausal coordination
takes place only when necessary –i.e., counterexpectational but, when not in combination
with the polarity ellipsis construction, can only be subclausal coordination without
ellipsis. 33 Arguably, this restriction can be reduced to an economy principle on structure
building, which forbids the inclusion of silent structure unless its presence is necessary for
independent reasons. 34
8 Conclusions
The primary goal of this article has been to argue that the semantic difference between
corrective and counterexpectational but translates into a different syntax for each of them.
The specific generalizations, stated in (10) and (68) above, are repeated here as (90) and
(91), respectively.
(90) The syntax of adversative coordination
a. Corrective but (sino) always requires its conjuncts to be full clauses.
b. Counterexpectational but (pero) allows its conjuncts to be smaller than clauses.
(91) Behaviour of counterexpectational ‘but’
Counterexpectational but coordination is subclausal coordination without ellipsis
unless the second conjunct contains an element associated to the CP area. In the
latter case, clausal coordination plus ellipsis obtains.
The data discussed throughout sections 3 and 5 suggest that these generalizations are
essentially correct. I would also like to call the reader’s attention to the pervasivity of ellipsis
in adversative coordination, which bears on the debate of how sentence fragments ought
to be analyzed. The analysis I have developed clearly favours the hypothesis that such fragments
are derived from an underlying full clausal structure, as claimed in Morgan (1973),
Merchant (2004a), and related works. Especially, the data discussed in section 3 pose a challenge
for the family of analyses where fragmentary sentences never have a hidden syntactic
structure (e.g., Culicover and Jackendoff 2005, Stainton 2006, Nykiel and Sag 2009, and
references).
Acknowledgements Many individuals deserve my gratitude for their comments on previous incarnations
of this paper. In reverse alphabetical order: Mark de Vos, Johan Rooryck, Jim McCloskey, Anikó Lipták,
Jorge Hankamer, Sandra Chung, the audiences at Tilburg, Utrecht, Deusto, Ottawa, and Leipzig, three anonymous
NLLT reviewers, and one non-anonymous NLLT editor (John Moore). For help with judgements, I’m
33 Thanks to John Moore (p.c.) for stressing the importance of this point.
34 See, e.g., the visibility guideline for functional categories discussed in Fukui and Sakai (2003:323–328).
This guideline states that functional heads are only justified (i) if they are overt; or (ii) in case they are silent,
if they have detectable side effects like affecting the morphology of another constituent or triggering some
movement operation. One might also want to include semantic effects as a third factor legitimating silent
structure, so as to be able to cover the bound pronoun cases in section 3.6.

On the syntax of adversative coordination 31
indebted to Mark de Vos, Anna McNay, Jim McCloskey, Robert Henderson, and Bill Haddican (English)
and Susana Huidobro, Ricardo Etxepare, Ángel Gallego, and Marcela Depiante (Spanish). Financial help has
been provided through pre- and post-doctoral grants from the Department of Education and Research of the
Basque Government (BFI03.277 and BFI07.32). All remaining errors and shortcomings are nobody’s fault
but mine.
References
Anscombre, Jean-Claude and Oswald Ducrot. 1977. Deux mais en français? Lingua 43:23–
40.
Barwise, Jon and Robin Cooper. 1981. Generalized quantifiers and natural language 4:159–
219.
Bianchi, Valentina and Roberto Zamparelli. 2004. Edge coordinations: focus and conjunction
reduction. In Peripheries: syntactic edges and their effects, eds. David Adger, Cécile
de Cat, and George Tsoulas, 313–328. Dordrecht: Kluwer.
Blackmore, Diane. 1989. Denial and contrast: a Relevance Theoretic account of but. Linguistics
& Philosophy 12:15–37.
Blackmore, Diane. 2000. Indicators and procedures: nevertheless and but. Journal of Linguistics
36:463–486.
Bosque, Ignacio. 1980. Sobre la negación. Madrid: Cátedra.
Chomsky, Noam. 1986. Barriers. Cambridge: MIT Press.
Cinque, Guglielmo. 1999. Adverbs and functional categories: a cross-linguistic perspective.
Oxford: Oxford University Press.
Culicover, Peter and Ray Jackendoff. 2005. Simpler syntax. Oxford: Blackwell.
Depiante, Marcela. 2000. The syntax of deep and surface anaphora: a study of null complement
anaphora and stripping/bare argument ellipsis. Doctoral dissertation, University of
Connecticut, Storrs.
Depiante, Marcela. 2004. Dos casos de elipsis con partícula de polaridad en español: evidencia
a favor de una visión no uniforme de la elipsis. Revista de la Sociedad Argentina
de Lingüística 1:53–69.
Drubig, Hans Bernhard. 1994. Island constraints and the syntactic nature of focus and
association with focus. Arbeitspapiere des SFB 340, n. 51, Tübingen.
Evans, Gareth. 1980. Pronouns. Linguistic Inquiry 11:337–362.
von Fintel, Kai. 1994. Restrictions on quantifier domains. Doctoral dissertation, University
of Massachusetts, Amherst.
Fox, Danny. 1999. Reconstruction, binding theory, and the interpretation of chains. Linguistic
Inquiry 30:157–196.
Fukui, Naoki and Hiromu Sakai. 2003. The visibility guideline for functional categories:
verb raising in japanese and related issues. Lingua 113:321–375.
Gallego, Ángel. 2004. Minimalist edge coordinations. Ms., Universitat Autonoma de
Barcelona.
Grice, Paul. 1975. Logic and conversation. In Syntax and semantics 3: Speech acts, eds.
Peter Cole and Jerry Morgan, 43–58. New York: Academic Press.
Hankamer, Jorge. 1973. Constraints on deletion in syntax. Doctoral dissertation, Yale University.
Horn, Laurence. 1989. A natural history of negation. Stanford: CSLI Publications.
Johannessen, Janne Bondi. 1998. Coordination. Oxford: Oxford University Press.
Kennedy, Christopher and Jason Merchant. 2000. Attributive comparative deletion. Natural
Language and Linguistic Theory 18:89–146.

32 Luis Vicente
van Koppen, Marjo. 2005. One probe, two goals: aspects of agreement in Dutch dialects.
Doctoral dissertation, Leiden University.
Laka, Itziar. 1990. Negation in syntax: on the nature of functional categories and projections.
Doctoral dissertation, MIT.
Lakoff, Robin. 1971. If’s, and’s and but’s about conjunction. In Studies in linguistic semantics,
eds. Charles Fillmore and D. Terence Langendoen, 114–149. New York: Holt.
Langacker, Ronald. 1969. On pronominalization and the chain of command. In Modern
studies in English, eds. Sanford Schane and David Reibel, 160–186. Englewood Cliffs,
NJ: Prentice-Hall.
Lasnik, Howard. 1972. Analyses of negation in English. Doctoral dissertation, MIT.
Lasnik, Howard. 2001. When can you save a structure by destroying it? In Proceedings of
NELS 31, eds. Minjoo Kim and Uri Strauss, 301–320. Amherst, Massachusetts: GLSA.
McCawley, James. 1991. Contrastive negation and metalinguistic negation. In CLS 27: the
parasession on negation, eds. Lise Dobrin, Lynn Nichols, and Rosa Rodríguez, 189–206.
Chicago: Chicago Linguistic Society.
Merchant, Jason. 2001. The syntax of silence: sluicing, islands, and the theory of ellipsis.
Oxford University Press, Oxford.
Merchant, Jason. 2004a. Fragments and ellipsis. Linguistics & Philosophy 27:661–738.
Merchant, Jason. 2004b. Remarks on stripping. Ms., University of Chicago.
Merchant, Jason. 2008. Variable island repair under ellipsis. In Topics in ellipsis, ed. Kyle
Johnson, 132–153. Cambridge: Cambridge University Press.
Morgan, Jerry. 1973. Sentence fragments and the notion of ‘sentence.’. In Issues in linguistics:
papers in honor of Henry and Renée kahane, eds. Braj Kachru, Robert Lees,
Yakov Malkiel, Angelina Pietrangeli, and Sol Saporta, 719–751. Urbana–Champaign:
UIUC Press.
Munn, Alan. 1993. Topics in the syntax and semantics of coordinate structures. Doctoral
dissertation, University of Maryland, College Park.
Nykiel, Joanna and Ivan Sag. 2009. Sluicing and stranding. Handout, 2009 LSA Annual
Meeting.
Ordóñez, Francisco. 1997. Word order and clause structure in Spanish and other Romance
languages. Doctoral dissertation, CUNY Graduate Center.
Progovac, Ljiljana. 1998a. Structure for coordination, Part I. GLOT International 3:3–6.
Progovac, Ljiljana. 1998b. Structure for coordination, Part II. GLOT International 3:3–9.
Reinhart, Tanya. 1991. Elliptical conjunctions: non–quantificational LF. In The chomskyan
turn, ed. Asa Kasher, 360–384. Oxford: Blackwell.
Repp, Sophie. 2005. Interpreting ellipsis: the changeable presence of negation in gapping.
Doctoral dissertation, Humboldt Universität, Berlin.
Rodrigues, Cilene, Andrew Nevins and Luis Vicente. 2009. Cleaving the interactions between
sluicing and preposition stranding. In Romance languages and linguistic theory
2006, eds. Danièle Torck and W. Leo Wetzels, 175–198. Amsterdam: John Benjamins.
Romero, Maribel. 1998. Focus and reconstruction effects in wh- phrases. Doctoral dissertation,
University of Massachusetts, Amherst.
Ross, John. 1967. Constraints on variables in syntax. Doctoral dissertation, MIT.
Ross, John. 1969. Guess who? In Proceedings of CLS 5, ed. Robert Binnick, 252–286.
Sag, Ivan, Gerald Gazdar, Tom Wasow and Steven Weisler. 1985. Coordination and how to
distinguish categories. Natural Language and Linguistic Theory 3:117–171.
Sauerland, Uli. 1996. Guess how? In Proceedings of ConSOLE 4, eds. João Costa, Rob
Goedemans, and Ruben van de Vijver, 297–309. Leiden: SOLE.

On the syntax of adversative coordination 33
Soltan, Usama. 2007. On formal feature licensing in minimalism: aspects of Standard Arabic
morphosyntax. Doctoral dissertation, University of Maryland, College Park.
Stainton, Robert. 2006. Words and thoughts: subsentences, ellipsis, and the philosophy of
language. Oxford: Oxford University Press.
Ticio, Emma. 2003. On the structure of DPs. Doctoral dissertation, University of Connecticut,
Storrs.
Toosarvandani, Maziar. 2009. Contrastive ‘but’ involves gapping not in Farsi but in English.
Handout, LSA 2009 Annual Meeting.
Umbach, Carla. 2005. Contrast and information structure: a focus-based analysis of but.
Linguistics 43:207–232.
Vicente, Luis. 2006. Short negative replies in Spanish. In Linguistics in the Netherlands 23,
eds. Jeroen van de Weijer and Bettelou Los, 199–210. Amsterdam: John Benjamins.
Vicente, Luis. 2008. Syntactic isomorphism and non-isomorphism under ellipsis. Ms.,
University of California, Santa Cruz.
van der Wouden, Ton. 1997. Negative contexts. Doctoral dissertation, University of Groningen.
Wurmbrand, Susi. 2008. Nor: neither disjunction nor paradox. Linguistic Inquiry 39:511–
522.
Zeijlstra, Hedde. 2004. Sentential negation and negative concord. Doctoral dissertation,
University of Amsterdam.